Krystalografia
Joanna Sołomiewicz
15 listopada 2018
Podsumowanie analizy danych
W mojej analizie na samym początku przy wczytywaniu danych odrzuciłam atrybuty, które nie są przeze mnie używane w dalszej analizie, ani regresji bądź klasyfikacji.
Następnie z mojego zbioru usunęłam rekordy, w których wystąpiła niewiadoma wartość. Zdecydowałam się na usunięcie, ponieważ zbiór danych jest na tyle duży, że mogłam sobie na to pozwolić zamiast zastępowania tych wartości średnimi, co jest pewnego rodzaju przekłamaniem.
Dodatkowo ze zbioru pozbyłam się tych ligandów, które zostały uwzględnione w zbiorze do usunięcia.
W kolejnym punkcie wykonałam krótkie podsumowanie najważniejszych danych wykorzystywanych w dalszym przetwarzaniu.
Następnie znalazłam 50 najczęściej występujących ligandów i ograniczyłam zbiór właśnie do nich oraz przedstawiłam je na histogramie, posortowane według częstości wystąpienia.
Jednym z wyzwań było przedstawienie korelacji między zmiennymi. Chciałam w tym celu użyć biblioteki corrplot, która przedstawia graficznie macierz korelacji. Niestety przy tak dużej ilości atrybutów musiałam z tego zrezygnować i postanowiłam przedstawić korelacje w postaci tabeli dodatkowo ograniczając wyniki tylko do tych, gdzie korelacja jest większa bądź równa 0.999.
Aby zaprezentować rozkład wartości liczby atomów i elektronów zastosowałam wykresy gęstościowe.
Dziesięć największych niezgodności liczby atomów i elektronów przedstawiłam w formie tabeli, podając dla każdej grupy ligandów uśrednioną różnicę.
Kolejnym z wyzwań było zaprezentowanie rozkładu wartości kolumn part_01. Wynikało to z ilości tych kolumn. W celu drobnej optymalizacji czytelności wykresów zastosowałam krok usunięcia outlier’ów oraz odcięcia prefixu “part_01_” z nazw kolumn, ponieważ występował w każdej z nich. Początkowo planowałam zastosować bibliotekę ggplot2, niestety nie dawała rady nawet przy 10 tysiącach rekordów. Następnie całość przepisałam na bibliotekę plotly, co zadziałało dla 10 tysięcy rekordów, ale niestety dla całości danych wygenerowało raport wielkości ponad 0.4 GB, którego nie byłam w stanie otworzyć w przeglądarce. Ostatecznie byłam zmuszona użyć najprostszego rozwiązania ze statycznymi wykresami.
Regresję liniową wykonałam dla standardowych parametrów. Stratyfikowany podział zbioru na uczący i testowy jako odpowiednio 70% i 30% danych oraz wykorzystanie funkcji lm.
Tworząc klasyfikator zmieniłam stosunek podziału na zbiór uczący i testowy, ponieważ danych było wystarczająco dużo. Dlatego ostatecznie stratyfikowany podział wynosi 99% danych dla zbioru uczącego i 1% dla zbioru testowego. Do uczenia użyłam algorytmu Random Forest z parametrem ntrees = 1000.
Wykorzystane biblioteki
library(dplyr)
library(ggplot2)
library(plotly)
library(tidyr)
library(knitr)
library(caret)
library(data.table)
library(tibble)Wczytywanie danych
removable_columns <- c("title", "pdb_code", "res_id", "chain_id", "local_res_atom_count", "local_res_atom_non_h_occupancy_sum", "local_res_atom_non_h_electron_occupancy_sum", "local_res_atom_C_count", "local_res_atom_N_count", "local_res_atom_O_count", "local_res_atom_S_count", "dict_atom_C_count", "dict_atom_N_count", "dict_atom_O_count", "dict_atom_S_count", "skeleton_data", "skeleton_cycle_4", "skeleton_diameter", "skeleton_cycle_6", "skeleton_cycle_7", "skeleton_closeness_006_008", "skeleton_closeness_002_004", "skeleton_cycle_3", "skeleton_avg_degree", "skeleton_closeness_004_006", "skeleton_closeness_010_012", "skeleton_closeness_012_014", "skeleton_edges", "skeleton_radius", "skeleton_cycle_8_plus", "skeleton_closeness_020_030", "skeleton_deg_5_plus", "skeleton_closeness_016_018", "skeleton_closeness_008_010", "skeleton_closeness_018_020", "skeleton_average_clustering", "skeleton_closeness_040_050", "skeleton_closeness_014_016", "skeleton_center", "skeleton_closeness_000_002", "skeleton_density", "skeleton_closeness_030_040", "skeleton_deg_4", "skeleton_deg_0", "skeleton_deg_1", "skeleton_deg_2", "skeleton_deg_3", "skeleton_graph_clique_number", "skeleton_nodes", "skeleton_cycles", "skeleton_cycle_5", "skeleton_closeness_050_plus", "skeleton_periphery", "fo_col", "fc_col", "weight_col", "grid_space", "solvent_radius", "solvent_opening_radius", "part_step_FoFc_std_min", "part_step_FoFc_std_max", "part_step_FoFc_std_step")
data <- fread("./all_summary.csv", nrows = 10000, header = TRUE, drop = removable_columns)
dim(data)## [1] 10000 350Przetwarzanie brakujących danych
dim(data)## [1] 10000 350data <- data %>%
drop_na()
dim(data)## [1] 8958 350Usuwanie niepotrzebnych ligandów
deletable_res_name <- c("UNK", "UNX", "UNL", "DUM", "N", "BLOB", "ALA", "ARG", "ASN", "ASP", "CYS", "GLN", "GLU", "GLY", "HIS", "ILE", "LEU", "LYS", "MET", "MSE", "PHE", "PRO", "SEC", "SER", "THR", "TRP", "TYR", "VAL", "DA", "DG", "DT", "DC", "DU", "A", "G", "T", "C", "U", "HOH", "H20", "WAT")
data <- data %>% filter(!res_name %in% deletable_res_name)
dim(data)## [1] 8910 350Podsumowanie danych
cols_summary <- c("res_name", "local_res_atom_non_h_count", "local_res_atom_non_h_electron_sum", "dict_atom_non_h_count", "dict_atom_non_h_electron_sum")
statistics <- data %>%
select(one_of(cols_summary))
kable(summary(statistics))| res_name | local_res_atom_non_h_count | local_res_atom_non_h_electron_sum | dict_atom_non_h_count | dict_atom_non_h_electron_sum | |
|---|---|---|---|---|---|
| Length:8910 | Min. : 1.0 | Min. : 7.00 | Min. : 1.00 | Min. : 7.0 | |
| Class :character | 1st Qu.: 4.0 | 1st Qu.: 31.00 | 1st Qu.: 4.00 | 1st Qu.: 31.0 | |
| Mode :character | Median : 6.0 | Median : 48.00 | Median : 6.00 | Median : 48.0 | |
| NA | Mean :13.3 | Mean : 99.63 | Mean : 13.76 | Mean :102.9 | |
| NA | 3rd Qu.:19.0 | 3rd Qu.:136.00 | 3rd Qu.: 20.00 | 3rd Qu.:141.0 | |
| NA | Max. :91.0 | Max. :727.00 | Max. :104.00 | Max. :727.0 |
dim(data)## [1] 8910 35050 najpopularniejszych ligandów
popular_ligands <- data %>%
select(res_name) %>%
count(res_name, sort = TRUE) %>%
slice(1:50)
popular_names_vector <- popular_ligands %>%
pull(res_name)
data <- data %>% filter(res_name %in% popular_names_vector)
dim(data)## [1] 6239 350Korelacja między zmiennymi
data %>%
select_if(is.numeric) %>%
cor() %>%
as.data.frame() %>%
rownames_to_column() %>%
gather(rowname2, value, -rowname) %>%
filter(value >= 0.9999, rowname != rowname2)## rowname rowname2 value
## 1 part_00_max local_max 1.0000000
## 2 part_01_max local_max 1.0000000
## 3 part_02_max local_max 1.0000000
## 4 part_00_max_over_std local_max_over_std 1.0000000
## 5 part_01_max_over_std local_max_over_std 1.0000000
## 6 part_02_max_over_std local_max_over_std 0.9999999
## 7 part_00_density_segments_count part_00_shape_segments_count 1.0000000
## 8 part_00_shape_segments_count part_00_density_segments_count 1.0000000
## 9 part_00_shape_M000 part_00_volume 1.0000000
## 10 part_00_density_M000 part_00_electrons 1.0000000
## 11 local_max part_00_max 1.0000000
## 12 part_01_max part_00_max 1.0000000
## 13 part_02_max part_00_max 1.0000000
## 14 local_max_over_std part_00_max_over_std 1.0000000
## 15 part_01_max_over_std part_00_max_over_std 1.0000000
## 16 part_02_max_over_std part_00_max_over_std 0.9999999
## 17 part_00_volume part_00_shape_M000 1.0000000
## 18 part_01_density_FL part_00_density_FL 0.9999229
## 19 part_01_density_I4 part_00_density_I4 0.9999031
## 20 part_00_electrons part_00_density_M000 1.0000000
## 21 part_01_density_segments_count part_01_shape_segments_count 1.0000000
## 22 part_01_shape_segments_count part_01_density_segments_count 1.0000000
## 23 part_01_shape_M000 part_01_volume 1.0000000
## 24 part_01_density_M000 part_01_electrons 1.0000000
## 25 local_max part_01_max 1.0000000
## 26 part_00_max part_01_max 1.0000000
## 27 part_02_max part_01_max 1.0000000
## 28 local_max_over_std part_01_max_over_std 1.0000000
## 29 part_00_max_over_std part_01_max_over_std 1.0000000
## 30 part_02_max_over_std part_01_max_over_std 0.9999999
## 31 part_01_volume part_01_shape_M000 1.0000000
## 32 part_00_density_FL part_01_density_FL 0.9999229
## 33 part_00_density_I4 part_01_density_I4 0.9999031
## 34 part_01_electrons part_01_density_M000 1.0000000
## 35 part_02_density_segments_count part_02_shape_segments_count 1.0000000
## 36 part_02_shape_segments_count part_02_density_segments_count 1.0000000
## 37 part_02_shape_M000 part_02_volume 1.0000000
## 38 part_02_density_M000 part_02_electrons 1.0000000
## 39 local_max part_02_max 1.0000000
## 40 part_00_max part_02_max 1.0000000
## 41 part_01_max part_02_max 1.0000000
## 42 local_max_over_std part_02_max_over_std 0.9999999
## 43 part_00_max_over_std part_02_max_over_std 0.9999999
## 44 part_01_max_over_std part_02_max_over_std 0.9999999
## 45 part_02_volume part_02_shape_M000 1.0000000
## 46 part_02_electrons part_02_density_M000 1.0000000Liczność najpopularniejszych ligandów według nazwy
plot_ligands <- ggplot(popular_ligands, aes(x = reorder(res_name, -n), y = n, fill = n)) +
geom_bar(stat = "identity") +
theme(axis.text.x = element_text(angle = 90)) +
xlab("ligand")+
ylab("liczność") +
labs(title = "Liczność ligandów według nazwy")
ggplotly(plot_ligands)Rozkłady gęstościowe liczb
Atomów
plot_atom <- ggplot(data, aes(x = local_res_atom_non_h_count)) +
geom_density(alpha = .3, fill = "#00CECB", color = NA) +
xlab("liczność atomów") +
ylab("gęstość") +
labs(title = "Rozkład gęstościowy atomów")
ggplotly(plot_atom)Elektronów
plot_electron <- ggplot(data, aes(x = local_res_atom_non_h_electron_sum)) +
geom_density(alpha = .3, fill = "#FF5E5B", color = NA) +
xlab("liczność elektronów") +
ylab("gęstość") +
labs(title = "Rozkład gęstościowy elektronów")
ggplotly(plot_electron)Największe niezgodności liczby
Atomów
data %>%
select(res_name, local_res_atom_non_h_count, dict_atom_non_h_count) %>%
group_by(res_name) %>%
summarise(atom_inconsistency = mean(abs(local_res_atom_non_h_count - dict_atom_non_h_count))) %>%
arrange(-atom_inconsistency) %>%
slice(1:10) %>%
kable()| res_name | atom_inconsistency |
|---|---|
| PLC | 17.1481481 |
| LHG | 4.4615385 |
| C8E | 2.6428571 |
| NDP | 1.7333333 |
| NAP | 1.5090909 |
| PG4 | 1.4225352 |
| MLY | 1.2222222 |
| CME | 1.0000000 |
| MAN | 1.0000000 |
| NAG | 0.9949495 |
Elektronów
data %>%
select(res_name, local_res_atom_non_h_electron_sum, dict_atom_non_h_electron_sum) %>%
group_by(res_name) %>%
summarise(electron_inconsistency = mean(abs(local_res_atom_non_h_electron_sum - dict_atom_non_h_electron_sum))) %>%
arrange(-electron_inconsistency) %>%
slice(1:10) %>%
kable()| res_name | electron_inconsistency |
|---|---|
| PLC | 114.444444 |
| LHG | 34.096154 |
| C8E | 16.714286 |
| NDP | 11.333333 |
| NAP | 10.654545 |
| PG4 | 9.633803 |
| MLY | 9.370370 |
| CME | 8.000000 |
| MAN | 8.000000 |
| NAG | 7.959596 |
Rozkład wartości kolumn part_01
remove_outliers <- function(data, na.rm = TRUE, ...) {
qnt <- quantile(data, probs=c(.25, .75), na.rm = na.rm, ...)
iqr <- 1.5 * IQR(data, na.rm = na.rm)
data_no_outliers <- data
data_no_outliers[data < (qnt[1] - iqr)] <- NA
data_no_outliers[data > (qnt[2] + iqr)] <- NA
data_no_outliers[!is.na(data_no_outliers)]
data_no_outliers
}
plot_part_data <- data %>%
select(contains("part_01"))
dim(plot_part_data)## [1] 6239 106plot_part_data <- plot_part_data %>%
sapply(remove_outliers) %>%
as.data.frame() %>%
drop_na()
names(plot_part_data) <- gsub("part_01_", "", names(plot_part_data))
plot_part_data <- plot_part_data %>%
gather(part, value, 1:106)
dim(plot_part_data)## [1] 320968 2ggplot(plot_part_data, aes(x = part, y = value)) +
geom_boxplot() +
facet_wrap(~part, scales = "free", ncol = 6) +
labs(title = "Rozkład wartości kolumn part_01")
Regresja liniowa
Liczba atomów
data_partition <- data %>%
select_if(is.numeric)
set.seed(111)
partition <- createDataPartition(
y = data_partition$local_res_atom_non_h_count,
p = .7,
list = FALSE)
data_train <- data_partition %>%
slice(partition)
data_test <- data_partition %>%
slice(-partition)
dim(data_train)## [1] 4368 347dim(data_test)## [1] 1871 347set.seed(111)
fit <- train(local_res_atom_non_h_count ~ ., data = data_train, method = "lm")
fit## Linear Regression
##
## 4368 samples
## 346 predictor
##
## No pre-processing
## Resampling: Bootstrapped (25 reps)
## Summary of sample sizes: 4368, 4368, 4368, 4368, 4368, 4368, ...
## Resampling results:
##
## RMSE Rsquared MAE
## 2.405387 0.9278705 0.1214908
##
## Tuning parameter 'intercept' was held constant at a value of TRUEset.seed(111)
prediction <- predict(fit, newdata = data_test)
postResample(pred = prediction, obs = data_test$local_res_atom_non_h_count)## RMSE Rsquared MAE
## 0.58514089 0.99823080 0.07083429Liczba elektronów
data_partition <- data %>%
select_if(is.numeric)
set.seed(111)
partition <- createDataPartition(
y = data_partition$local_res_atom_non_h_electron_sum,
p = .7,
list = FALSE)
data_train <- data_partition %>%
slice(partition)
data_test <- data_partition %>%
slice(-partition)
dim(data_train)## [1] 4369 347dim(data_test)## [1] 1870 347set.seed(111)
fit <- train(local_res_atom_non_h_electron_sum ~ ., data = data_train, method = "lm")
fit## Linear Regression
##
## 4369 samples
## 346 predictor
##
## No pre-processing
## Resampling: Bootstrapped (25 reps)
## Summary of sample sizes: 4369, 4369, 4369, 4369, 4369, 4369, ...
## Resampling results:
##
## RMSE Rsquared MAE
## 33.47963 0.8446019 1.274699
##
## Tuning parameter 'intercept' was held constant at a value of TRUEset.seed(111)
prediction <- predict(fit, newdata = data_test)
postResample(pred = prediction, obs = data_test$local_res_atom_non_h_electron_sum)## RMSE Rsquared MAE
## 2.6911609 0.9991266 0.4644615Klasyfikator
Przewidywanie wartości res_name
dim(data)## [1] 6239 350removable_columns <- c("blob_coverage", "res_coverage", "local_res_atom_non_h_count", "local_res_atom_non_h_electron_sum", "dict_atom_non_h_count", "dict_atom_non_h_electron_sum")
data_partition <- data %>%
select(-removable_columns)
dim(data_partition)## [1] 6239 344data_partition$res_name <- as.factor(data_partition$res_name)
set.seed(111)
partition <- createDataPartition(
y = data_partition$res_name,
p = .7,
list = FALSE)
data_train <- data_partition %>%
slice(partition)
data_test <- data_partition %>%
slice(-partition)
dim(data_train)## [1] 4391 344dim(data_test)## [1] 1848 344set.seed(111)
fit <- train(
res_name ~ .,
data = data_train,
method = "rf",
ntree = 10,
na.action = na.pass)
fit## Random Forest
##
## 4391 samples
## 343 predictor
## 50 classes: 'ACT', 'ACY', 'ADP', 'AMP', 'BR', 'C8E', 'CA', 'CD', 'CIT', 'CL', 'CME', 'CU', 'CYC', 'DMS', 'EDO', 'EPE', 'FAD', 'FE', 'FE2', 'FEC', 'FES', 'FMN', 'FMT', 'GLC', 'GOL', 'H4B', 'HEC', 'HEM', 'IOD', 'K', 'LHG', 'MAN', 'MES', 'MG', 'MLY', 'MN', 'MPD', 'NAD', 'NAG', 'NAP', 'NDP', 'NI', 'PEG', 'PG4', 'PGE', 'PLC', 'PLP', 'PO4', 'SO4', 'ZN'
##
## No pre-processing
## Resampling: Bootstrapped (25 reps)
## Summary of sample sizes: 4391, 4391, 4391, 4391, 4391, 4391, ...
## Resampling results across tuning parameters:
##
## mtry Accuracy Kappa
## 2 0.3910456 0.3411316
## 172 0.5329791 0.4972496
## 343 0.5309613 0.4952548
##
## Accuracy was used to select the optimal model using the largest value.
## The final value used for the model was mtry = 172.set.seed(111)
prediction <- predict(fit, newdata = data_test)
confusionMatrix(data = prediction, data_test$res_name)## Confusion Matrix and Statistics
##
## Reference
## Prediction ACT ACY ADP AMP BR C8E CA CD CIT CL CME CU CYC DMS EDO EPE
## ACT 6 0 0 0 0 0 0 0 0 2 1 0 0 0 5 0
## ACY 0 4 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## ADP 0 0 5 0 0 0 0 0 0 0 0 0 0 0 0 0
## AMP 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0
## BR 0 0 0 0 4 0 1 0 0 0 0 0 0 0 0 0
## C8E 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0
## CA 0 0 0 0 0 0 59 2 0 3 0 1 0 0 0 0
## CD 0 0 0 0 0 0 1 10 0 0 0 0 0 0 0 0
## CIT 1 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0
## CL 0 0 0 0 4 0 7 2 0 89 0 0 0 0 0 0
## CME 0 0 0 0 0 0 0 0 0 0 4 0 0 0 0 0
## CU 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 0
## CYC 0 0 0 0 0 0 0 0 0 0 0 0 4 0 0 0
## DMS 1 0 0 0 0 0 0 0 0 0 0 0 0 72 4 1
## EDO 8 1 0 0 0 1 1 0 2 2 1 1 1 5 76 1
## EPE 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 4
## FAD 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## FE 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0
## FE2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## FEC 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## FES 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
## FMN 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0
## FMT 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## GLC 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
## GOL 9 3 2 3 0 2 0 0 1 0 0 0 0 3 28 1
## H4B 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## HEC 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## HEM 0 0 1 0 0 1 0 0 0 0 0 0 0 0 2 0
## IOD 0 0 0 0 0 0 3 0 0 2 0 0 0 0 0 0
## K 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0
## LHG 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0
## MAN 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## MES 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0
## MG 0 0 0 0 1 0 8 2 0 11 0 0 0 0 1 0
## MLY 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## MN 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0
## MPD 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
## NAD 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## NAG 1 0 1 1 0 0 0 0 0 0 1 0 1 0 2 0
## NAP 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0
## NDP 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## NI 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
## PEG 0 0 0 0 0 0 0 0 0 0 1 0 0 0 3 0
## PG4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0
## PGE 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0
## PLC 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## PLP 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 2
## PO4 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0
## SO4 8 3 2 5 0 1 0 1 0 0 0 0 1 18 14 3
## ZN 0 0 0 0 0 0 2 7 0 3 0 4 0 0 0 0
## Reference
## Prediction FAD FE FE2 FEC FES FMN FMT GLC GOL H4B HEC HEM IOD K LHG MAN
## ACT 0 0 0 0 0 0 2 0 5 0 0 3 0 0 0 0
## ACY 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## ADP 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0
## AMP 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## BR 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## C8E 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## CA 0 0 0 0 0 0 0 0 0 0 0 0 1 12 0 0
## CD 0 0 1 0 1 0 0 0 0 0 0 0 0 2 0 0
## CIT 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## CL 0 0 0 0 0 0 0 0 0 0 0 0 5 3 0 0
## CME 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
## CU 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## CYC 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## DMS 0 0 0 0 0 0 1 0 2 0 0 0 0 0 0 0
## EDO 2 0 0 1 0 0 3 0 26 0 1 7 1 0 0 0
## EPE 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## FAD 14 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0
## FE 0 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## FE2 0 0 8 0 0 0 0 0 0 0 0 0 0 0 0 0
## FEC 0 0 0 4 0 0 0 0 1 0 0 0 0 0 0 0
## FES 0 0 0 0 5 0 0 0 0 0 0 0 0 0 0 0
## FMN 1 0 0 0 0 11 0 0 0 0 0 0 0 0 0 0
## FMT 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## GLC 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## GOL 2 0 0 1 0 2 3 3 101 0 0 6 0 0 3 4
## H4B 1 0 0 0 0 1 0 0 0 5 0 0 0 0 0 0
## HEC 0 0 0 0 0 0 0 0 0 0 5 0 0 0 0 0
## HEM 2 0 0 1 0 0 0 0 1 0 4 42 0 0 0 1
## IOD 0 0 0 0 0 0 0 0 0 0 0 0 18 0 0 0
## K 0 0 0 0 0 0 0 0 0 0 0 0 0 4 0 0
## LHG 0 0 0 0 0 0 0 0 0 0 0 0 0 0 12 0
## MAN 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3
## MES 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
## MG 0 1 0 0 0 0 0 0 0 0 0 0 2 1 0 0
## MLY 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0
## MN 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## MPD 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
## NAD 1 0 0 0 0 0 0 0 1 1 0 2 0 0 0 0
## NAG 1 0 0 0 0 0 0 1 1 1 0 1 0 0 0 1
## NAP 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## NDP 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
## NI 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## PEG 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0
## PG4 2 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0
## PGE 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## PLC 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## PLP 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0
## PO4 0 0 0 0 1 1 0 0 3 0 0 0 0 0 0 0
## SO4 4 0 0 0 0 1 0 1 27 0 0 7 1 0 0 0
## ZN 0 3 3 0 1 0 0 0 0 0 0 0 0 0 0 0
## Reference
## Prediction MES MG MLY MN MPD NAD NAG NAP NDP NI PEG PG4 PGE PLC PLP PO4
## ACT 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
## ACY 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0
## ADP 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## AMP 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0
## BR 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## C8E 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## CA 0 14 0 10 0 0 0 0 0 0 0 0 0 0 0 0
## CD 0 0 0 1 0 0 0 0 0 2 0 0 0 0 0 0
## CIT 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
## CL 0 7 0 0 0 0 0 0 0 1 1 0 0 0 0 0
## CME 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0
## CU 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## CYC 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## DMS 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1
## EDO 0 2 1 0 1 0 3 0 0 0 7 1 2 0 0 1
## EPE 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0
## FAD 0 0 0 0 0 3 0 0 0 0 0 0 0 0 1 0
## FE 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## FE2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## FEC 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## FES 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## FMN 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## FMT 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## GLC 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0
## GOL 2 1 3 0 4 10 14 3 3 0 15 7 1 0 2 7
## H4B 0 0 0 0 0 0 2 0 1 0 0 0 0 0 0 0
## HEC 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## HEM 0 0 0 0 0 1 1 0 0 0 0 0 0 0 0 2
## IOD 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0
## K 0 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0
## LHG 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## MAN 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## MES 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## MG 0 28 0 2 0 1 0 0 0 1 0 0 0 0 0 0
## MLY 0 0 7 0 0 0 0 0 0 0 1 0 0 0 0 0
## MN 0 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0
## MPD 0 0 0 0 0 0 1 0 0 0 0 1 0 0 0 0
## NAD 0 0 0 0 1 17 3 0 3 0 0 0 0 0 1 0
## NAG 1 0 2 0 0 4 87 0 0 0 2 3 3 0 2 1
## NAP 0 0 0 0 0 0 0 9 2 0 0 0 0 0 0 0
## NDP 0 0 0 0 0 0 1 1 2 0 0 0 0 0 0 0
## NI 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0
## PEG 1 0 0 0 0 1 0 0 0 0 2 2 1 0 0 0
## PG4 0 0 0 0 0 0 1 0 0 0 2 3 0 0 0 0
## PGE 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
## PLC 0 0 0 0 0 0 0 0 0 0 0 0 0 8 0 0
## PLP 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 1
## PO4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 10
## SO4 0 0 2 0 2 1 3 3 1 0 1 2 1 0 1 27
## ZN 0 1 0 1 0 0 0 0 0 2 0 0 0 0 0 0
## Reference
## Prediction SO4 ZN
## ACT 1 0
## ACY 0 0
## ADP 0 0
## AMP 0 0
## BR 0 0
## C8E 0 0
## CA 0 9
## CD 0 6
## CIT 0 0
## CL 0 2
## CME 1 0
## CU 0 0
## CYC 0 0
## DMS 3 0
## EDO 16 0
## EPE 0 0
## FAD 0 0
## FE 1 2
## FE2 0 1
## FEC 0 0
## FES 0 0
## FMN 0 0
## FMT 1 0
## GLC 0 0
## GOL 24 0
## H4B 0 0
## HEC 0 0
## HEM 1 0
## IOD 0 0
## K 1 1
## LHG 0 0
## MAN 1 0
## MES 0 0
## MG 0 6
## MLY 0 0
## MN 0 2
## MPD 0 0
## NAD 1 0
## NAG 1 0
## NAP 0 0
## NDP 0 0
## NI 0 1
## PEG 1 0
## PG4 0 0
## PGE 0 0
## PLC 0 0
## PLP 0 0
## PO4 11 0
## SO4 230 1
## ZN 0 64
##
## Overall Statistics
##
## Accuracy : 0.5703
## 95% CI : (0.5474, 0.5931)
## No Information Rate : 0.1591
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.5361
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: ACT Class: ACY Class: ADP Class: AMP Class: BR
## Sensitivity 0.176471 0.363636 0.416667 0.000000 0.444444
## Specificity 0.988975 0.999456 0.999455 0.998910 0.999456
## Pos Pred Value 0.230769 0.800000 0.833333 0.000000 0.800000
## Neg Pred Value 0.984632 0.996202 0.996200 0.992958 0.997287
## Prevalence 0.018398 0.005952 0.006494 0.007035 0.004870
## Detection Rate 0.003247 0.002165 0.002706 0.000000 0.002165
## Detection Prevalence 0.014069 0.002706 0.003247 0.001082 0.002706
## Balanced Accuracy 0.582723 0.681546 0.708061 0.499455 0.721950
## Class: C8E Class: CA Class: CD Class: CIT Class: CL
## Sensitivity 0.250000 0.70238 0.384615 0.250000 0.78070
## Specificity 1.000000 0.97052 0.992316 0.998913 0.98155
## Pos Pred Value 1.000000 0.53153 0.416667 0.500000 0.73554
## Neg Pred Value 0.996750 0.98561 0.991228 0.996746 0.98552
## Prevalence 0.004329 0.04545 0.014069 0.004329 0.06169
## Detection Rate 0.001082 0.03193 0.005411 0.001082 0.04816
## Detection Prevalence 0.001082 0.06006 0.012987 0.002165 0.06548
## Balanced Accuracy 0.625000 0.83645 0.688466 0.624457 0.88112
## Class: CME Class: CU Class: CYC Class: DMS Class: EDO
## Sensitivity 0.500000 0.333333 0.571429 0.73469 0.54676
## Specificity 0.997826 1.000000 1.000000 0.99200 0.94207
## Pos Pred Value 0.500000 1.000000 1.000000 0.83721 0.43429
## Neg Pred Value 0.997826 0.996748 0.998373 0.98524 0.96234
## Prevalence 0.004329 0.004870 0.003788 0.05303 0.07522
## Detection Rate 0.002165 0.001623 0.002165 0.03896 0.04113
## Detection Prevalence 0.004329 0.001623 0.002165 0.04654 0.09470
## Balanced Accuracy 0.748913 0.666667 0.785714 0.86335 0.74442
## Class: EPE Class: FAD Class: FE Class: FE2 Class: FEC
## Sensitivity 0.333333 0.424242 0.555556 0.666667 0.571429
## Specificity 0.999455 0.997245 0.997281 0.999455 0.998914
## Pos Pred Value 0.800000 0.736842 0.500000 0.888889 0.666667
## Neg Pred Value 0.995659 0.989612 0.997824 0.997825 0.998371
## Prevalence 0.006494 0.017857 0.004870 0.006494 0.003788
## Detection Rate 0.002165 0.007576 0.002706 0.004329 0.002165
## Detection Prevalence 0.002706 0.010281 0.005411 0.004870 0.003247
## Balanced Accuracy 0.666394 0.710744 0.776418 0.833061 0.785171
## Class: FES Class: FMN Class: FMT Class: GLC
## Sensitivity 0.625000 0.687500 0.0000000 0.000000
## Specificity 0.999457 0.998362 0.9994562 0.997827
## Pos Pred Value 0.833333 0.785714 0.0000000 0.000000
## Neg Pred Value 0.998371 0.997274 0.9951272 0.996204
## Prevalence 0.004329 0.008658 0.0048701 0.003788
## Detection Rate 0.002706 0.005952 0.0000000 0.000000
## Detection Prevalence 0.003247 0.007576 0.0005411 0.002165
## Balanced Accuracy 0.812228 0.842931 0.4997281 0.498914
## Class: GOL Class: H4B Class: HEC Class: HEM
## Sensitivity 0.57386 0.625000 0.500000 0.60000
## Specificity 0.89713 0.997283 1.000000 0.98988
## Pos Pred Value 0.36996 0.500000 1.000000 0.70000
## Neg Pred Value 0.95238 0.998368 0.997287 0.98434
## Prevalence 0.09524 0.004329 0.005411 0.03788
## Detection Rate 0.05465 0.002706 0.002706 0.02273
## Detection Prevalence 0.14773 0.005411 0.002706 0.03247
## Balanced Accuracy 0.73550 0.811141 0.750000 0.79494
## Class: IOD Class: K Class: LHG Class: MAN Class: MES
## Sensitivity 0.64286 0.181818 0.800000 0.333333 0.285714
## Specificity 0.99670 0.996714 0.999454 0.999456 0.998914
## Pos Pred Value 0.75000 0.400000 0.923077 0.750000 0.500000
## Neg Pred Value 0.99452 0.990207 0.998365 0.996746 0.997289
## Prevalence 0.01515 0.011905 0.008117 0.004870 0.003788
## Detection Rate 0.00974 0.002165 0.006494 0.001623 0.001082
## Detection Prevalence 0.01299 0.005411 0.007035 0.002165 0.002165
## Balanced Accuracy 0.81978 0.589266 0.899727 0.666395 0.642314
## Class: MG Class: MLY Class: MN Class: MPD Class: NAD
## Sensitivity 0.48276 0.437500 0.0588235 0.000000 0.447368
## Specificity 0.97933 0.998362 0.9978154 0.997826 0.992265
## Pos Pred Value 0.43077 0.700000 0.2000000 0.000000 0.548387
## Neg Pred Value 0.98317 0.995103 0.9913185 0.995662 0.988442
## Prevalence 0.03139 0.008658 0.0091991 0.004329 0.020563
## Detection Rate 0.01515 0.003788 0.0005411 0.000000 0.009199
## Detection Prevalence 0.03517 0.005411 0.0027056 0.002165 0.016775
## Balanced Accuracy 0.73104 0.717931 0.5283195 0.498913 0.719817
## Class: NAG Class: NAP Class: NDP Class: NI Class: PEG
## Sensitivity 0.73729 0.562500 0.153846 0.1428571 0.064516
## Specificity 0.98150 0.998362 0.997820 0.9989136 0.993396
## Pos Pred Value 0.73109 0.750000 0.333333 0.3333333 0.142857
## Neg Pred Value 0.98207 0.996187 0.994028 0.9967480 0.984188
## Prevalence 0.06385 0.008658 0.007035 0.0037879 0.016775
## Detection Rate 0.04708 0.004870 0.001082 0.0005411 0.001082
## Detection Prevalence 0.06439 0.006494 0.003247 0.0016234 0.007576
## Balanced Accuracy 0.85940 0.780431 0.575833 0.5708854 0.528956
## Class: PG4 Class: PGE Class: PLC Class: PLP
## Sensitivity 0.142857 0.0000000 1.000000 0.166667
## Specificity 0.996169 0.9994565 1.000000 0.996732
## Pos Pred Value 0.300000 0.0000000 1.000000 0.250000
## Neg Pred Value 0.990207 0.9956687 1.000000 0.994565
## Prevalence 0.011364 0.0043290 0.004329 0.006494
## Detection Rate 0.001623 0.0000000 0.004329 0.001082
## Detection Prevalence 0.005411 0.0005411 0.004329 0.004329
## Balanced Accuracy 0.569513 0.4997283 1.000000 0.581699
## Class: PO4 Class: SO4 Class: ZN
## Sensitivity 0.200000 0.7823 0.67368
## Specificity 0.988877 0.9086 0.98460
## Pos Pred Value 0.333333 0.6183 0.70330
## Neg Pred Value 0.977998 0.9566 0.98236
## Prevalence 0.027056 0.1591 0.05141
## Detection Rate 0.005411 0.1245 0.03463
## Detection Prevalence 0.016234 0.2013 0.04924
## Balanced Accuracy 0.594438 0.8455 0.82914